Strategic Decision or Structural Advantage?

Open original

Using Databricks token-share data, Tom Tunguz argues that Anthropic and OpenAI have split the enterprise AI market: Anthropic dominates premium frontier models while OpenAI wins the cheap, high-volume tier. The open question is whether OpenAI's aggressive Luna price cuts are a temporary share grab or a real cost advantage in its architecture.

  • Anthropic's share of frontier-tier token volume rose from 81% in July to 94% by November 2, after the Claude Opus 5.5 launch.
  • OpenAI's low-end share jumped from 30% to 81% after repeated Luna price cuts, totaling a 90% reduction from the original price.
  • In the mid tier, OpenAI went from 0% to 47% of volume in about four months, taking 57% of the newest Sonnet 5.5 versus Sol cohort within six weeks of launch.
  • Open-weight model share grew from 7% to 33% by September, then stalled right when OpenAI cut prices again.

Anthropic and OpenAI each hold roughly 90% of a segment, but the segments differ. Anthropic owns the premium end, with 94% of state-of-the-art token volume by November 2, 2026, up from 81% in July after Claude Opus 5.5 launched on September 21. OpenAI dominates the low end at 81%, with its economy models (GPT-6 Luna, GPT-5 Nano, and older models) carrying 19.3% of all proprietary tokens. The author's question is whether this split reflects a deliberate pricing strategy or a structural cost gap in how each lab serves tokens. The data comes from Databricks' enterprise customer base, which skews toward data-platform and agentic-coding work, so it is not a census of the whole market. The mid tier is where the two labs now collide. OpenAI went from 0% of Sonnet-versus-Sol mid-tier volume in July to 47% by November 2, and captured 57% of the newest cohort (Sonnet 5.5 against GPT-6 Sol and 6.1 Sol) within six weeks of launch. OpenAI got there through price. It cut Luna by 80% on July 30, taking input and output pricing from $1.00 and $6.00 to $0.20 and $1.20 per million tokens, then launched GPT-6 Luna permanently at $0.10 and $0.50 on September 22, a further 50% below promotional pricing. Luna's low-end share rose from 30% to 66% after the first cut and peaked at 84%. OpenAI attributed part of the room to cut prices to a 20% reduction in serving cost from GPU kernel work and a 15%-plus gain in token throughput from speculative decoding, which bears on the structural-cost question. The open question is whether Luna's pricing is a temporary subsidy to buy share or evidence of a lasting cost edge. The author also points to a signal worth watching: open-weight token share climbed every week from 7% in July to 33% in September, then stalled at 31% in the same week the 50% cuts landed. That stall suggests price cuts from the closed labs can blunt open-weight momentum, though one week of data does not settle it. The post's authors read the numbers carefully, but the cost-structure question depends on figures neither lab has published in full, so the answer is still open.

Ask about this article